Utilizing an Artificial Intelligence and Machine Learning Model to Predict Colorectal Cancer Risk in American Samoa:...

This pilot study aimed to assess the feasibility of using an artificial Intelligence and machine learning  (AI/ML) model to predict colorectal cancer (CRC) risk in American Samoa, where resource limitations and cultural barriers significantly hinder screening efforts.

The AI/ML model used complete blood count (CBC) results, along with age and gender, to predict CRC risk. A retrospective analysis was conducted on data from 6,025 individuals aged 50 and above from the Lyndon Baines Johnson Tropical Medical Center's electronic health records. Of these, 62 participants were identified as high-risk for CRC based on the AI/ML model. The study also incorporated the methylated Septin 9 (mSept9) biomarker as an alternative, less invasive screening method for CRC detection. Participants were contacted for follow-up CRC screening, which included colonoscopy, fecal immunochemical testing (FIT), or mSEPT9 blood testing.

The AI/ML model identified 62 high-risk participants. However, only four participants returned for further testing, and just one agreed to a colonoscopy. The colonoscopy result revealed a benign polyp and low hemoglobin levels in the participant with the highest risk score. mSEPT9 levels were elevated in this participant, indicating the potential utility of this biomarker for early CRC detection. Despite promising results, the model's validation was limited due to low participation in follow-up screening.

This study demonstrates the potential of AI/ML models for predicting CRC risk in resource-limited and culturally diverse populations like American Samoa. However, significant barriers, including cultural, financial, and logistical factors, limit patient follow-up and the broader implementation of these technologies. Future research should focus on addressing these barriers, enhancing community engagement, and integrating culturally appropriate interventions to improve CRC screening and outcomes in underserved populations.